Modern scientific cosmology pushes the boundaries of knowledge and the knowable. This is prompting questions on the nature of scientific knowledge, and the emergence of the new field “Philosophy of Cosmology.” One central issue is what defines a “good” model. Conventional methods for choosing good models have a drawback in data-sparse situations: enabling the implicit introduction of value judgments, which can determine inference and lead to inferential polarization, e.g., on the question of ultimate explanation. Therefore, additional dimensions for comparing models are needed.

Evaluation of scientific evidence is based on statistical methods, which rely on comparing what is observed with what is not observed. Performing this comparison tells us what is probable, and what is not. However, when addressing the global properties of the Universe such as its initial state, or its fundamental existence, we have only one single “outcome” of an “experiment.” We cannot generate new universes in an experiment to test our theories in the conventional way. Thus, we are fundamentally limited in our knowledge about what is not observed (e.g., parts of the Universe we cannot see, or non-existence), and hence how to assess the probability of what is observed. Indeed, it is not clear that the notion of probability has any meaning in this context. Still, the question of whether the Universe is probable is very important in our current scientific context.

'Fine-tuning of the Universe' is the claim that a number of parameters in our current fundamental physical theories must take on values within a narrow range, to allow for the development of the type of macroscopic structures that we see in the Universe – and ultimately our own existence. Is the Universe fine-tuned for life? In light of the above remarks on probability, fine-tuning cannot be regarded as simple fact. Rather, it is a specific interpretation of available evidence in a domain where the interpretative framework is unproven or questionable, and possibly untestable. Indeed, when we speak of the Universe being fine-tuned, we compare our Universe to some hypothetical collection of universes and judge that our Universe is “special.” Even if we do not explicitly talk about a multiverse, we cannot avoid implicitly referring to some such collection – regardless of whether it truly exists or not. How can we know what this kind of collection looks like? What are the prospects for testing our hypotheses about it? If we have not addressed those questions, is there any meaning in talking about “fine-tuning of the Universe”?

To address this scenario, I have proposed a three-legged model comparison approach: evidence, elegance and beneficence (see here, and in 'The Philosophy of Cosmology', Cambridge University Press 2017). This explicitly considers the categories of criteria that are always at least implicitly used by practising scientists. I have presented a tentative model comparison framework on this basis, which extends the Bayesian statistical framework. The practical implementation now remains to be worked out.